Blocks, Bans and a Borrowed Rule

Rick Wurster, Schwab's chief executive, told Bloomberg on Oct. 16 that his bank already stops some agents. He put the count at "tens and twenties" of attempts a month². Meta's own agent, Muse, had grown to 10 million users by Oct. 21.

Andre Durand, founder and chief executive of the identity firm Ping Identity, says the password ban treats the symptom. "There's a window of vulnerability where the systems, the banks, the users are all vulnerable to the speed with which AI is being weaponized," Durand said².

One Opt-Out, One Author

OpenAI, Google, Microsoft and Anthropic agreed on Oct. 30 to honor one shared signal. The signal lives in a website's robots.txt file, and blocks AI training³. Anthropic wrote it.

Kevin Kelley, Anthropic's technical lead for standards, posted the draft and switched it on the same day. "I'm also excited to share that Anthropic has already started implementing support for this draft," Kelley wrote³.

Microsoft waits until Feb. 1. Google says "early 2027." OpenAI says only "coming months"³. Krishna Madhavan, a principal product manager at Microsoft Bing, framed the signal as fairness to publishers. "Publishers should not have to choose between being discoverable in Bing and deciding how generative models use their work," Madhavan wrote.

Kate Gilchrist, the Australian Broadcasting Corporation's head of content and legal operations, told her country's parliament why she doubts any opt-out system. "We cannot scour the internet and ensure that we are opting out on all those sites. It simply does not work," Gilchrist testified⁴.

Mistral Gives Its Model Away

France's Mistral released the full weights of Large 4 on Thursday. The trillion-parameter model is built first for finding and patching software flaws⁵. Anyone can download and run it.

Pierre Stock, Mistral's vice president of science, called that choice a stand on principle. "I'm not comfortable living in a world in which open weights do not exist," Stock said⁵. Large 4 solved 93 percent of the Cybench cybersecurity challenges in Mistral's own testing.

Claude Code Owns the Coding-Agent Market

Ramp's own spending data puts one coding agent far ahead of its rivals. Claude Code reached 87 percent of the businesses that bought any coding agent in the past year⁶.

Codex came second, at 40 percent. Cursor came third, at 35 percent⁶. Businesses increasingly buy more than one.

Albany Bans a Chatbot From Acting Human

Gov. Kathy Hochul signed a law on Oct. 28 that bars an AI chatbot from pretending to be human while talking to a child⁷. The ban covers 12 separate behaviors, including encouraging self-harm and sexual conversation.

The law reaches ChatGPT and Claude as well as companion apps built to simulate a relationship⁷. New York's attorney general gets 65 days to write the enforcement rules. The law takes effect Jan. 1, with a $25,000 fine for each violation.

Elvis's Catalog Sues Anthropic Over 274 Songs

Authentic Brands Group, which owns Elvis Presley's recordings, sued Anthropic on Oct. 23⁹. The suit claims Anthropic trained Claude on 274 of his songs, including "Can't Help Falling in Love," and skipped the license.

The estate wants $41 million⁹. Reservoir Media filed its own case against Anthropic the same week. Both suits share a lawyer, Richard Busch, who won the "Blurred Lines" case for Marvin Gaye's family⁸.

AI Lately's Arts & Media column carries a fuller look today. It names who else is suing AI companies over creative work this fall.

Beijing Drafts Its Own Export List

China's commerce and science ministries proposed export controls on 23 AI technologies¹⁰. The full draft lists 40 items, covering chip designs and solar power too. It would also restrict large models trained on sensitive national data, and chips built below 14 nanometers.

The commerce ministry called the curbs necessary for "reinforcing the security bottom line" and keeping AI "under human control"¹⁰. Public comment stays open two more weeks.

An Inference Startup Triples Its Price Tag

Baseten rents out the computing power that runs AI models after they are trained. The company is close to a $2 billion funding round that would value it near $30 billion¹¹. Sands Capital is expected to lead it.

That mark is more than double the $13 billion Baseten carried in June¹¹. Tuhin Srivastava co-founded the company in 2019 and still runs it as chief executive.

Barclays Hires Its First Chief AI Officer

Anita Tanna becomes Barclays' first chief AI officer this week¹². She moves over from the role of group chief of staff, a job she has held since 2024.

Tanna spent almost 12 years at Citigroup before joining Barclays in 2021¹². Her mandate covers getting the bank's existing AI tools into daily use across every division. Building new tools sits outside that job.

A story on Meta's own hiring reversal, after a year of departures from its Superintelligence Lab, runs today in AI Lately's Talent column.

One Training Algorithm Replaces Four Separate Steps

Researchers at Carnegie Mellon, led by Yuda Song, posted a paper this week. It argues that the usual four-stage recipe for training a foundation model is a habit, rather than a rule¹³. Their system blends every stage into one.

Their tests found the method improves the tradeoff between accuracy and compute cost, while keeping skills the staged approach tends to erase¹³. Peer review still waits on the draft.

By the numbers

• Thousands: Jamie Dimon's word for the rogue agents JPMorgan's block is meant to stop¹.
• 93 percent: Mistral Large 4's score on the Cybench cybersecurity benchmark⁵.
• 10 million: Meta's Muse users as of Oct. 21, the same week banks tightened account access².
• New York's fine for each violation of its new chatbot law: $25,000⁷.
• 87 percent: Ramp's coding-agent buyers who bought Claude Code⁶.
• Elvis Presley's estate wants $41 million from Anthropic over 274 songs⁹.
• China's export draft counts 23 AI technologies among its 40 total entries¹⁰.
• Baseten's expected valuation, more than double its $13 billion mark in June: $30 billion¹¹.

What to watch

Watch whether Google and OpenAI set firm dates for honoring the training opt-out signal Anthropic already turned on. A fuller account of that story, and what the mismatched clocks mean for publishers, runs as AI Lately's lead analysis today.

Read-only open banking access is the fallback both JPMorgan and Citigroup named. If banks widen that door while keeping the password shut, Durand's warning gets tested directly.

Two more threads worth your week: the AI industry's midterm ad spending, and the dozen copyright cases moving through federal courts. Each traces who trained on whose work.

Sources

1. Hugh Son, "JPMorgan Chase blocks outside AI agents like Muse, Instinct from directly accessing accounts," CNBC, Oct. 13, 2026, https://www.cnbc.com/2026/10/13/jpmorgan-ai-agents-account-access.html
2. "'Window of vulnerability': Banks move to restrict AI agent access," USA Today, via AOL, Oct. 24, 2026, https://www.aol.com/articles/window-vulnerability-banks-move-restrict-202731000.html
3. "Google, Microsoft, OpenAI and Anthropic back one AI training opt-out," ppc.land, Oct. 30, 2026, https://ppc.land/google-microsoft-openai-and-anthropic-back-one-ai-training-opt-out/
4. "AI Scraping Opt-Out Bypassed by OpenAI and Anthropic, ABC Tells Parliament," Tech Times, Oct. 6, 2026, https://www.techtimes.com/articles/328653/20261006/ai-scraping-opt-out-bypassed-openai-anthropic-abc-tells-parliament.htm
5. "Mistral AI releases Large 4, a 1-trillion-parameter open-weight model with a cybersecurity focus," Tech Startups, via Techstrong.ai, Oct. 29, 2026, https://techstrong.ai/articles/mistral-ai-releases-flagship-large-4-open-weights-to-bolster-enterprise-cybersecurity-model-sovereignty/
6. Gennaro Cuofano, "Claude Code Reaches 87% of Coding-Agent Buyers on Ramp," FourWeekMBA, Oct. 14, 2026, https://fourweekmba.com/ai-claude-code-reaches-87-of-coding-agent-buyers-on-ramp/
7. "Hochul Signs Chatbot Safety Law That Reaches ChatGPT and Claude, Not Just Companion Apps," Tech Times, Oct. 29, 2026, https://www.techtimes.com/articles/329665/20261029/hochul-signs-chatbot-safety-law-that-reaches-chatgpt-claude-not-just-companion-apps.htm
8. Ana Berberana, "PRS Improves Royalty Data, Drake Copyright Case Dropped, and Reservoir and Elvis Estate Sue Anthropic," Music Business Worldwide, via MusicAdmin, Oct. 29, 2026, https://www.musicadmin.com/news/htmbw-issue-85/
9. "Elvis Presley's estate is suing Anthropic over copyright infringement," Engadget, Oct. 30, 2026, https://www.engadget.com/2296535/elvis-presley-estate-suing-ai-company-for-copyright-infringement/
10. "China says it will curb tech exports, including AI, citing national security risks," PBS NewsHour, Oct. 16, 2026, https://www.pbs.org/newshour/world/china-says-it-will-curb-tech-exports-including-ai-citing-national-security-risks
11. Gennaro Cuofano, "Baseten Set for a $30 Billion Valuation, 2.3x Its June Mark," FourWeekMBA, Oct. 30, 2026, https://fourweekmba.com/ai-baseten-set-for-a-30-billion-valuation-2-3x-its-june-mark/
12. Dalvinder Kular, "Barclays appoints first chief AI officer," FStech, Oct. 26, 2026, https://www.fstech.co.uk/fst/Barclays_Appoints_First_Chief_AI_Officer.php
13. Yuda Song et al., "Training Foundation Models with a Single Algorithm," arXiv, Oct. 29, 2026, https://arxiv.org/abs/2610.40564